Re: [PATCH 07/22] tests/acceptance/virtiofs_submounts.py: evaluate string not length

[Philippe Mathieu-Daudé]

On 2/9/21 1:52 PM, Alex Bennée wrote:
> Max Reitz <mreitz@redhat.com> writes:
>> On 09.02.21 12:24, Alex Bennée wrote:
>>> Max Reitz <mreitz@redhat.com> writes:
>>>> On 04.02.21 14:23, Alex Bennée wrote:
>>>>> Cleber Rosa <crosa@redhat.com> writes:
>>>>>
...
>>>> Ideally, vmlinuz='' would be the default, but there’s a problem with
>>>> that: I submitted this test along with the patches that added the
>>>> required feature to the Linux kernel, so at that point that feature was
>>>> not part of Linux.  That’s why you generally have to point it to a Linux
>>>> kernel binary you built yourself that has this feature (5.10 does).
>>>
>>> This is where it deviates from the rest of the check-acceptance tests.
>>> Generally I don't have to worry about finding the right image myself.
>>
>> Yes, but there’s nothing I can do apart from just not having the test as 
>> part of qemu, which I don’t find better than to just cancel it when not 
>> run with the necessary parameters.
> 
> No I agree having the tests is useful.
> 
>>
>> Or, well, I could let the test download and compile the Linux sources, 
>> but I don’t think that’s a viable alternative.
> 
> There has been discussion before but I agree it's not viable given the
> compile times for such things.
> 
>>> At the least it would be worth pointing to a part of our docs on how
>>> to build such an image.
>>
>> Well, I could perhaps come up with a comprehensive kernel configuration 
>> that works, but I really don’t think that’s valuable for people who just 
>> want to run the acceptance tests and don’t care about this test in 
>> particular.  I just don’t think they’re actually going to build a Linux 
>> kernel just for it.
> 
> Sure - but I was trying to review the series and as part of my review I
> generally like to run the things I review. Hence why I stopped as I
> couldn't get things running.
> 
>> (Alternatively, I could just build a Linux kernel here on my machine, 
>> upload the binary and point to it somewhere, e.g. in the cancel message. 
>>   That would be OK for me, and perhaps something I could imagine someone 
>> might actually use.)
> 
> I've actually done this with some Xen patches I'm working on at the
> moment. I'll probably decorate the test with:
> 
>   @skipUnless(os.getenv('AVOCADO_ALLOW_UNTRUSTED_CODE'), 'untrusted code')
> 
> with a comment explaining what's waiting to be upstreamed. Once there
> are upstream binaries I plan on transitioning the test to those.

Instead of a binary AVOCADO_ALLOW_UNTRUSTED_CODE variable, we could
have a list allowed domains/namespaces, that can be increased on the
tester discretion.

For example these are assumed trusted:

. archives.fedoraproject.org
. archive.debian.org
. cdn.netbsd.org
. github.com/torvalds
. people.debian.org/~aurel32
. snapshot.debian.org
. storage.kernelci.org
. www.qemu-advent-calendar.org

Then personally interested in testing ARM boards I'd amend:

. apt.armbian.com
. github.com/philmd
. github.com/groeck
. github.com/hskinnemoen
. github.com/pbatard

and Max's repo since I'm interested in testing virtiofs_submounts.